Color = how unusual the face looks to radar right now: the face's 2026 season-mean morning-radar anomaly (the scalable screening statistic), ranked against every face-season in this region's 2020–25 history. Red does not mean "will collapse"; it means "darker to morning radar than this region's faces almost ever are", which is the signal that preceded the Langtang collapse.
Numbered pins = where anomaly meets consequence. Each face also carries three fixed factors: how far a failure could fall, whether standing water sits below it, and how many people live in the valleys within 50 km. The pins mark the top faces by anomaly × consequence, in the same order as the list below.
Click any face for its year-by-year history and its consequence factors. Where basin data is loaded, clicking also lights up the face's downstream watershed chain (HydroBASINS, pre-linked, no flow modelling), labelled with the population living within ~1 km of the river channel in each basin (GHSL masked to a MERIT-Hydro channel corridor), so the route from face to people is visible. Consequence never changes a face's color; it only decides which anomalies deserve a human first.
Each polygon is one computed monitoring unit (steep glacier
face) from the exploratory 0005 enumeration. Color places the face's 2026 season-mean morning-radar anomaly (the scalable screening statistic) as a percentile within the region's own
2020–25 history of the same statistic; it is a rank against the past, not a
prediction. Consequence factors (fall height, standing water below the face, valley
population) are radius-buffer proxies, not flow-path models; they rank
attention and never alter the anomaly tier itself. Snapshot dashboard,
refreshed by rerunning live_extract.py + live_build.py. Evidence base for the
method: two historical case studies, jointly p ≈ 0.02 — details in the
companion report.
Exploratory analysis, not an operational OCHA product or warning.